159

Chapter 6

QUALITY ADJUSTMENT OF PRICE INDICES IN INFORMATION ANDCOMMUNICATION TECHNOLOGY INDUSTRIES: SIMULATION OF EFFECTS ON

MEASURED REAL OUTPUT IN FIVE OECD COUNTRIES

by

Paul SchreyerDirectorate for Science, Technology and Industry, OECD

Introduction

The problems associated with the construction of price indices for goods whose specification andquality change rapidly have long been recognised by statisticians and economists: if qualityimprovements are inadequately captured, price indices will tend to overstate inflationarydevelopments and understate quantity changes. Perhaps the best documented example of the effectsof quality adjustments on the resulting price and quantity series is the case of computers: in theUnited States, the introduction of hedonic methods resulted in a deflator for computers and peripheralequipment that declined at an annual average rate of over 12 per cent during the 1980s, significantlyfaster than the conventional measure of price changes.

Everything else equal, such quality adjustments reduce drastically the measures of producerprices in the computer industry while raising its measure of real gross and net output. The size of theeffect of quality adjustments of computer prices on measures of output and productivity of industriesthat use these computers as investment goods or as intermediate products is, however, less obvious, asis the effect on measured aggregate net output, GDP.

While computers are the best-known example for products with rapid quality change, they are byno means the only example. Quite different products, for example the output of the constructionindustry, have also undergone quality changes that traditional price indices captured inadequately.1

Closer to the product nature of computers are other information and communication technology-related products such as telephone systems and services, semiconductors or automatic banking. Thecurrent chapter focuses on the effects of quality changes in those industries that are either producersof information and communication technologies (ICT) or heavy users of them.

We identify two major producer industries of ICT: the office machinery and computer industry;and the radio, TV and communication equipment industry. The case for considering the effects ofquality changes in the computer industry has already been stated; the case for the communicationequipment industry arises from the fact that this industry’s activities include, among other goods, theproduction of semiconductors. In terms of the user industries, we take a look at the inter-industrydistribution of ICT investment flows (Table 1) to state that the communication services industry and

160

the banking and finance sector are those industries whose output is most clearly marked by the use ofICT capital goods and their services.

Table 1. Investment in information and communication technologyShares of sectors in total IT and ICT investment,1 percentages, 1990

Information technology Information and communication technologyFinance, insurance,banking, real estate

Finance, insurance,banking, real estate

Transport and communication services

Canada 18.5 10.5 40.6France 74.0 49.9 15.9Germany 44.4 .. ..Japan 47.8 35.1 14.7United Kingdom 52.2 31.8 22.3United States 44.0 29.3 22.1

1. Information technology (IT) investment is defined as the flow of capital goods from the office machinery and computerindustry to other sectors. Information and communication technology (ICT) investment includes also the capital goods flowsfrom the radio, TV and communication equipment industry.Source: Motohashi (1996).

With these four ICT-related industries identified, this chapter addresses the following threequestions. Supposing that output prices of ICT-related industries are being quality adjusted, how doesthis affect:

♦ Measured real value added in ICT-related industries themselves? We note that not only grossoutput prices may be quality adjusted, prices of intermediate inputs may be as well. What is thenet effect on measures of real value added?

♦ Measured real value added in other industries2? Downstream industries of the ICT-relatedsectors use intermediate ICT inputs. Under quality adjustment, the measured quantities of theseintermediate inputs will rise and reduce measured real value added in downstream industries.By how much? Which are the most affected industries?

♦ Measures of aggregate real value added? Quality adjustment of prices in ICT-related industrieswill probably raise the measured real value added of these industries but reduce the measuredreal value added in downstream industries – what is the net effect on GDP? Is it sizeable?What are the issues associated with aggregation of such effects?

These questions have a double focus. First, they all have to do with real value-added – themeasure we wish to look at as we are interested in the links between different sectors and betweensectoral and aggregate output. Second, real value added provides the link to labour productivitymeasures: as measures of labour input are independent of any quality adjustments of output prices,any observed effects on measures of real value added directly translate into effects on measures oflabour productivity, if the latter is measured in terms of real value added per hour worked or peremployed person.

This paper proceeds as follows: the following section provides a non-technical, and empiricaldescription of the approach adopted to provide some answers to the above questions; the third sectiongives a more formal presentation of the approach and presents results; and the final section presentsconclusions.

161

Quality adjustment: sectoral and aggregate effects

ICT producers and users

For the current purpose of tracing the effects of the quality adjustment of prices in ICT-relatedindustries on measures of sectoral and aggregate real value added, we start with a broad division ofindustries in three groups.

The first group, which we shall call major ICT producers, comprises the Office Machinery andComputer Industry (ISIC Rev.2, sector 3825) and the Radio, TV and Communication EquipmentIndustry (ISIC Rev.2, sector 3832). Practical considerations3 led us to consider only the activities ofthese two industries although other relevant ICT products may well remain hidden in othermanufacturing or services industries. However, we believe that these two industries capture thelarger part of relevant ICT products including computers, computer processors, peripherialequipment, semiconductors, telephones, faxes, etc.

The second group, which we shall call major ICT users, comprises the Communication ServicesIndustry (ISIC Rev.2, sector 72) and Banking and Insurance (ISIC Rev.2, sectors 81 and 82). Thesetwo industries have in common that the quality and diversity of their output have changed rapidly andthat this change is intimately associated with the use of ICT capital goods in the production process.It has been mentioned before that data on inter-industry flows of investment goods show (Table 1)that these two industries have in fact been the biggest acquirers of capital goods from the major ICTproducers. The third group consists simply of all other industries.

This leads us to observe that there are two distinct occurrences of rapid quality change related toICT, a point that is best shown with the help of Figure 1: in the upper left corner we show theproduction of ICT goods. These goods (e.g. computer processors or semiconductors) haveexperienced rapid quality improvements so there is a case of potential under-measurement of realgross output, unless output prices are quality adjusted. Often, the output of these ICT-producerindustries consists of investment goods (e.g. mainframe or desktop computers), widely sold to majorICT-user industries where they become part of the capital stock and where improved capital servicesenter the production process of the ICT-user industries. These improved capital inputs (say, the useof computers for telephone or banking services) have also led to rapid quality change in the output ofICT-user industries and here is the second case of potential under-measurement of real gross output,unless prices are quality adjusted.4

These two basic cases of quality change form the starting point for our analysis. To see howquality adjustments in prices of the four main ICT producers and users affect measurement of realvalue added in other industries, we need to trace the flow of intermediate inputs from the former tothe latter. This marks an important distinction: while investment goods are the link between mainICT producers and main ICT users, flows of intermediate goods need to be traced to find out how realvalue added measures of other industries could be affected. While computers are the most obviousexample for ICT investment goods, semiconductors, disks or CPUs are examples of intermediategoods used by downstream industries such as car manufacturing5. This, again, is schematised inFigure 1: flows of intermediate inputs may be mis-measured if they come for major ICT-producer or-user industries. Note that there are also mutual flows of intermediate inputs between major ICT-producer and major ICT-user industries with a potential impact on the measurement of industry realvalue added.

162

Figure 1. Quality adjustment of ICT-prices and their effect on value-added: a stylisedpresentation

Major ICT Producers: Major ICT Users: Other industriesOffice machinery and computer ind. Communication servicesRadio, TV,communication equipm. Banking and insurance

GROSSOUTPUT

minus

INTERMEDIATEINPUTS

equals

VALUE-ADDED

Rapid quality improvement:

Under-measurement of

real gross output

Rapid quality improvement:

Under-measurement ofreal gross output

No rapid quality improvement:

Correct measurement of

real gross output

ICT investment

goodsProduction of:

Production of communication,

finance, insurance services

ICT capital services

Production of other goods and services

Real intermediate inputs:

Possible mis-measurement

Real intermediate inputs:

Possible mis-measurement

Real intermediate inputs:

Possible mis-measurement

Flows of intermediate inputs:Input-Output Relationships

Real value-added:Possible mis-measurement

Real value-added:Possible mis-measurement

Real value-added:Possible mis-measurement

ICT intermediate

goods

163

Assumptions about mis-measurement

Only a small number of countries construct hedonic price indices for the output of the computerindustry and even fewer have made adjustments in the construction of price indices for other ICT-related industries. One important reason is that the establishment of hedonic indices is costly; otherreasons are of a more conceptual nature, for example, problems in identifying the appropriate unit ofoutput in the banking industry; yet another reason is empirical difficulties in identifying andmeasuring the characteristics necessary for the establishment of hedonic indices. Given this situation,any assessment of the effects of quality adjustment on measures of real output has to rely onsimulations of effects along a reasonable range of parameters. For the four ICT-related industries wechose the following parameter values:

♦ Office machinery and computer industry (ISIC Rev. 2, sector 3825). This is the only sector forwhich several countries have elaborated hedonic price indices (see Wyckoff, 1995) for anoverview). On average, and for the period under consideration (1985-90), quality-adjustedprices of computer processors and peripherial equipment fell at an annual rate of about 12 percent while unadjusted prices remained roughly constant. Sichel and Oliner (1994) report aannual average rate of change of -8.5 per cent for office, computing and accounting equipmentgoods in the United States. Office, computing and accounting equipment is an aggregate that,in addition to computers and peripherial equipment, includes typewriters and other non-computer office equipment and appears close to the definition of ISIC sector 3825. To keepthings simple, and in line with these orders of magnitude, we set the deviation in the rate ofchange between adjusted and unadjusted price indices to 10 percentage points.

♦ Radio, TV and communication equipment industry (ISIC Rev.2, sector 3832). This sector isanother important producer of information technology, mainly because its products includesemiconductors that may have experienced a quality-adjusted price decline not too far from thefall in computer prices. Flamm (1993) presents price indices for dynamic random accessmemory chips and for erasable programmable read-only memory chips which confirm thispoint. Flamm’s Fisher-type price index oscillates between -35 and +25 per cent during thesecond half of the 1980s, on average about -8 per cent. However, these products are only partof the entire sector’s activities which extend to a number of other products (e.g. radio and TVsets) whose quality changed much less radically. At the same time, these items appear to play aminor role in the entire industry: Canadian data (Statistics Canada, 1996) show, for examplethat in 1994, record player, radio and TV set production generated value added of C$ 0.1 billion(1986 Canadian dollars) while communications and other electronic equipment accounted forsome C$ 2.9 billion (1986 Canadian dollars). To allow, nevertheless, for products other thansemiconductors, we assume a deviation of quality-adjusted from unadjusted prices of5 percentage points per year. The exception to this assumption is Germany for which theavailable input-output tables do not permit sector 3832 to be separated – this sector is includedin its parent industry, sector 383. To be able to include Germany in our simulations, we have toadjust sector 383 instead of 3832. However, to account for other products in sector 383 thathave not undergone rapid quality improvements, we adjust its output price only by 2 per cent.

♦ Communication services (ISIC Rev. 2, sector 72). This sector, which includestelecommunications and postal services (but not broadcasting), counts among the mostsignificant users of information technology. While difficult to measure, more anecdotalevidence seems to support a plausible claim that the quality of communication services hasincreased rapidly but may not have been captured in volume indices of the communication

164

service industry. We assume a 2 percentage point difference between adjusted and unadjustedprice indices.

♦ Finance and insurance (ISIC Rev. 2, sectors 81 and 82). The banking sector has been thesubject of several analyses concerning the measurement of its output. The choice of the unit ofproduction is fraught with conceptual difficulties and proxies include the size of currentdeposits, new credits accorded or number of portfolios handled.6 In other cases, industry outputis extrapolated based on labour input, with an assumed or no productivity change.7 It is quitelikely, therefore, that these measures only inadequately capture the number and quality offinancial services that developed over the past 15 years, including credit cards, automatic tellingmachines, increased speed of transactions or the development of inter-bank networks. Sichel(1994) reports estimates by Popkin (1992) on banking services and concludes that conventionalmeasures may under-estimate output growth by 1.3 percentage points a year. For insurance,Popkin presents different estimates. The average that maximises under-measurement is3.9 percentage points per year. Baily and Gordon (1988), in their discussion of US productivitygrowth in finance, insurance and real estate sectors, suggest a rough estimate of theunderstatement of around 2 per cent post-1973. For our purposes, we consider an under-measurement of 1 percentage point per year.

Aggregation and additivity

The impact on measured aggregate real output is a weighted average of the mis-measurement inreal output in individual industries. Obviously, the size of the aggregate effect depends on the sizeand evolution of measurement errors in individual industries and on the share of these industries inGDP. Griliches (1994) argues that both measurement errors and the share of “unmeasurable” sectorshave increased. The second effect (share of unmeasurable sectors) depends, however, on the degreeto which measurement biases in real gross output translate into value added which makes up GDP.While the value-added share of services is high and has been rising, it is also true that a relativelylarge part of service industry output consists of intermediate input to other industries. Thus, anadjustment of a service industry’s output price may have different impact on aggregate real outputthan the price adjustment of an industry with a relatively small share of deliveries to intermediatedemand.

This observation led, for example, Sichel (1994) to consider the share of services in finaldemand rather than in GDP by industry, which showed that the share of services in final demand wasmuch smaller than the share of service industries in GDP. Sichel concludes that measurement errorsdue to a sectoral shift toward services are small and could not explain any significant portion of theproductivity slowdown.

The same point had been made by Baily and Gordon (1988), who drew a clear distinctionbetween the measurement errors in industry productivity that potentially affect aggregate value addedand the measurement errors that only change the composition of real output. The first case ariseswhen industry output is mainly directed at final demand (the example given by Baily and Gordon isreal consumer purchases of air transportation); the second case arises when industry output ispredominantly directed at intermediate demand. Baily and Gordon cite mis-measurement of outputgrowth in the railroad freight industry as a pure industry phenomenon, since all of railroad freightoutput is an intermediate good. The input-output approach employed in the present chapter offers anatural way to account for the share of intermediate deliveries in service sector output. It also allowsthe effects on other industries and GDP to be traced via changes in real intermediate deliveries.

165

One important methodological point in the evaluation of the effects of quality adjustment onmeasures of real value added is whether additivity of real variables is imposed or not. Additivityimplies that levels of variables at constant prices form a consistent accounting system, for examplethat the sum of real value added by industry equals real GDP. Alternatively, a less restrictive methodcan be chosen that does not impose additivity and only ensures a coherent system of growth rates ofreal variables. Under this alternative formulation, growth rates are combined with nominal weights(for example, the growth rate of real GDP is derived as a weighted average of real industry growthrates with current price value added as weights), while under additivity growth rates are implicitlycombined with real weights. The distinction is important when relative prices change rapidlybetween the periods under consideration (see also Sichel, 1994). Take the example of the computerindustry with a fast decline in quality-adjusted prices and a corresponding rapid increase in realoutput. If the real growth rate of the computer industry is weighted with real shares, these shares willincrease as they reflect the rise in real output but not the decline in relative prices. Nominal shares,on the other hand, will capture both the rise in quantity and the fall in prices with an unknown neteffect. The real growth rate of the computer industry weighted with nominal shares will, therefore,have less of an impact on aggregate GDP than the same growth rate weighted with real shares. Eachmethod has its justification and interpretation8 (see below for a more technical discussion) – whatmatters here is the fact that the choice of the aggregation procedure has a sizeable impact on theoverall results of our simulations.

Model and results

Input-output relations and deflation

The model applied for our analysis is a standard, open, static input-output model, with data fromthe OECD Input-Output Database that was analysed for five OECD countries (Canada, Germany,Japan, the United Kingdom and the United States) for 1985 and 1990 or closest years. Throughoutthe analysis, we maintain Assumption No. 1: mis-measurement, due to rapid quality change, relatesonly to the splitting of nominal series into price and volume components, not to the nominal values.Current-price flows of intermediate goods, output and value added are considered correctly measured.

The basic flow relation in the input-output models states that each industry’s gross output is thesum of sales of intermediate inputs and final demand deliveries, as expressed in equation [1], whereXi stands for the current-price gross output of industry j, Zij for the flow of current price

intermediate inputs from industry i to industry j, yi for the current-price final demand for industry i’s

products, and aij are current-price input coefficients:

X a X y with a Z Xi ij j i ij ij jj

= + =∑ / [1]

In our treatment of this simple relationship, Assumption No. 2 needs to be pointed out: thenominal flows of imported goods are simply added to the flows of domestically produced goods.Although this is a common procedure in input-output analysis (Miller and Blair, 1985), it implies thatprice indices of imported goods are assumed to move at the same rate as price indices of domesticallyproduced goods. This is of relevance in the next step where a price index of industry i’s gross output,Pi,, is applied to intermediate goods flows and to deliveries to final demand.

The next step is the deflation of each industry’s output, independent of its destination.Assumption No. 3 behind this row-wise deflation (see also Figure 2) is that there exists a unique price

166

index for each industry that applies independently of the sector to which this industry delivers. Thus,the price index of deliveries by the computer industry destined for final demand (for example,investment demand) is assumed to be the same as the price index of those deliveries destined for otherindustries’ intermediate demand. Obviously, this could introduce a bias if investment andintermediate goods within the computer industry are of such a different nature that their price indicesvary at widely different rates and if the composition of deliveries varies widely across receivingindustries and final demand.

Figure 2. Input-output approachAdditivity imposed

Real final demand

Real gross output

To: industry1 2 .. .. N C+I+G+X-M

1 Z11/P1 Z12/P1 .. .. Z1N/P1 Y1/P1 X1/P1

From: 2 Z21/P2 Z22/P2 .. .. Z1N/P2 Y2/P2 X2/P2

industry .. .. .. .. .. .. .. ..

.. .. .. .. .. .. .. ..

N ZN1/PN ZN2/PN .. .. ZNN/PN YN/PN XN/PN

Real value added X1/P1- X2/P2- .. .. XN/PN- GDP

Σ Zi1/Pi Σ Zi2/Pi .. .. Σ ZiN/Pi

Real gross output X1/P1 X2/P2 XN/PN

The sectoral price indices that we apply for row-wise deflation are simple sectoral deflators ofgross industry output, derived from constant-price input-output table. They are thus fixed-weight(Paasche-type) deflators, normalised to a common base year – 1985. Thus, Assumption No. 4 statesthat the nature of these sectoral deflators (fixed-weight) – which enter our model as “unadjusted priceindices” does not fundamentally change the subsequent simulation results. Whether this is a severeconstraint or not rests on the assessment of the underlying aggregation problem: the moredisaggregated input-output industries are, the less of a potential problem there is (in the present case,the OECD input-output tables allow to distinguish between 36 industries). It is difficult to assess theseverity of this problem, although the short time-period that we consider (five years) might suggestthat during such a relatively constrained time span and within fairly homogenous industries, the biasfrom fixed-weight indices should not be too important.

Introducing adjusted prices

Next, we introduce two alternative price indices for deflation in the four ICT-related industries:

piu , a price index unadjusted for quality changes; and pi

a , a price index adjusted for quality changes.

For all other industries, the two price indices coincide. qiu and qi

a denote the resulting variables for

industry i’s unadjusted and adjusted gross output at constant prices [2]. The expression in [], which

represents a system of equations in real gross output, can be solved for values of qiu and qi

a 9.

167

q X p ap

pq

y

piu

i iu

ijju

iu j

u i

iu

j

= = +∑/ and q X p ap

pq

y

pia

i ia

ijja

ia j

a i

ia

j

= = +∑/ [2]

On the basis of the two sets of deflators, two sets of industry value added at constant prices arecalculated, based on double deflation. There are, however, several ways to go about doubledeflation.10 The distinction we draw here is between additivity and non-additivity of the levels of realvalue added.

Double deflation with additivity imposed. In this case, double deflation consists of deducting thevalue of intermediate inputs at constant prices from the value of gross output at constant prices. Inthe input-output model, this amounts to subtracting the sum of all intermediate inputs that industry jpurchases, from the same industry’s constant-price gross output (see the two vertical arrows inFigure 2). Equation [3] shows the so-obtained adjusted and unadjusted sectoral real value added,

VAju~ and VAj

a~. The expression for real value added combines real gross output, q j , with Nij

i∑ ,

the sum of all real intermediate inputs used by industry j. As before, the subscripts a and u help todistinguish between adjusted and unadjusted variables:

VA q N with N Z pju

ju

iju

iiju

ij iu~

/= − =∑ and VA q N with N Z pja

ja

ija

iija

ij ia~

/= − =∑ [3]

From equation [3], the total effect of quality adjustment on each industry’s growth rate of realvalue added is derived as:

∆ ∆ln~

ln~

VA VAja

ju− [4]

At the same time, a further transformation of [3] allows to show that additivity implicitlyamounts to applying real weights to growth rates of real output and intermediate inputs:differentiation of the expressions in[3] (leaving out the superscripts for simplicity) yields equation[5]. This continuous-time expression can be approximated by a discrete Törnqvist-type expression asshown in [6]:

VA

VA

q

VA

q

q

N

VA

N

Nj

j

j

j

j

j

ij

j

ij

iji

~&

~ ~&

~&

= − ∑ [5]

( ) ( )∆ ∆ ∆ln~ ~

( )~

( ) ln ~ ( ) ~ ( ) lnVA t q t Nj j j j ij iji

ij≅ + − +∑θ θ τ χ χ τ12

12 [6]

with ~

/~θ j j jq VA= , the ratio of real gross output to real value added and ~ /

~χ ij ij jN VA= , the ratios

of real intermediate inputs to real value-added. Thus, additivity implies aggregation of growth rateswith real weights.

Double deflation without additivity. Under this formulation, additivity is not assumed acrossconstant price variables and the derivation of growth rates starts from the current-price relationship(see, for example, Kuroda and Shimpo, 1991; or Sakurai et al., 1996) between gross output, value

added and intermediate inputs, where p jv is a price index of industry j's value added:

168

p VA p q p Njv

j j j i iji

= − ∑ [7]

Differentiation of this expression with respect to time and approximation of the continuous-timeformula by a discrete Törnqvist index formulation gives an expression akin to equation [6], with one

important difference: the weights θ j j j jv

jp q p VA= / and χ ij i ij jv

jp N p VA= / are expressed in

current prices:

( ) ( )∆ ∆ ∆ln ( ) ( ) ln ( ) ( ) lnVA t q t Nj j j j ij iji

ij≅ + − +∑θ θ τ χ χ τ12

12 [8]

With the distinction between adjusted and non-adjusted variables, this formulation leads to thefollowing expression for the overall effects of quality adjustment in ICT-related industries on sectoralgrowth rates of real value added [9]. This formulation also allows the decomposition of the totaleffect on measured real value added into a gross-output-related effect (first expression on the right-hand side of the equation) and an intermediate input effect (second expression on the right-hand sideof the equation):

( ) [ ]( ) [ ]

∆ ∆

∆ ∆

∆ ∆

ln ln

( ) ( ) ln ln

( ) ( ) ln ln

VA VA

t q q

t N N

ja

ju

j j ja

ju

ij ij ija

iju

− =

+ − −

+ −

θ θ τ

χ χ τ

12

12

[9]

Sectoral value added: results

The effects of the adjustment of gross output prices on real value added growth in the four ICT-related industries and its decomposition are depicted in Table 2. Several points are worth noting:

♦ Measured change in real value added rises by more than the assumed deviation of qualityadjusted from non-adjusted change in gross output prices (10, 5, 2 and 1 percentage pointsaccording to industries). Note that the weights that translate the change in measured gross

output, ~θ j and θ j (the ratio between gross output and value added) both exceed unity, thus

magnifying the gross output effect on measured real value added. The adjustment effect of realintermediate inputs enters with a negative sign, because quantities of ICT-related inputs hadbeen under-measured. The net effect on measured industry value added is, however, clearlypositive.

♦ A comparison of the measurement effects under additivity assumptions and without them showsa marked difference: throughout, the effects on real value added measures are less pronouncedwhen additivity is imposed. The reason lies in the difference between nominal and real weightsthat distinguishes the two approaches. Under additivity assumptions, the weights that link the

rate of change of real gross output with real value added, ~θ j , fall over time, while the nominal

weights θ j that characterise the absence of additivity remain roughly constant. Thus, in the

first case, changes in real gross output generate less of an effect on measures of real valueadded than in the second case, where weights hardly change. Although there is a counter effect

169

associated with the weights that link the rate of change of real intermediate inputs with realvalue-added, ( ~χ j falls more rapidly in absolute terms than χ j ), it does not outweigh the first.

Table 2. Effects of quality adjustment of gross output prices on measured real value addedDifference between average annual percentage changes of adjusted and unadjusted measures

Method: additivity imposed, 1985-90 or closest year, ICT-related industriesCanada Germany Japan United

KingdomUnitedStates

Office machinery and computersReal value added: total effect 16.3 16.3 18.4 14.5 14.4 gross output effect 21.9 20.9 17.9 14.7 26.4 intermediate input effect -5.6 -4.6 0.5 -0.2 -12.0Radio, TV and comm. equipmentReal value added: total effect 6.4 3.3 8.5 6.7 7.1 gross output effect 10.0 3.7 11.5 9.0 12.9 intermediate input effect -3.6 -0.4 -3.0 -2.3 -5.8Communication servicesReal value added: total effect 2.2 2.2 2.4 2.4 2.2 gross output effect 2.4 2.4 2.5 2.7 2.5 intermediate input effect -0.2 -0.2 -0.1 -0.3 -0.3Banking and insuranceReal value added: total effect 1.3 1.3 1.3 1.6 1.3 gross output effect 1.6 1.4 1.4 1.8 1.8 intermediate input effect -0.3 -0.1 -0.1 -0.2 -0.5

Table 3. Effects of quality adjustment of gross output prices on measured real value addedDifference between average annual percentage changes of adjusted and unadjusted measures

Method: no additivity imposed, 1985-90 or closest year, ICT-related industriesCanada Germany Japan United

KingdomUnitedStates

Office machinery and computersReal value added: total effect 18.1 22.5 24.0 20.9 20.8 gross output effect 29.7 27.0 31.6 27.7 27.7 intermediate input effect -11.6 -4.5 -7.6 -6.8 -6.9Radio, TV and comm. equipmentReal value added: total effect 7.2 3.4 11.3 8.9 10.4 gross output effect 11.6 4.2 15.8 12.3 14.0 intermediate input effect -4.4 -0.8 -4.5 -3.4 -3.6Communication servicesReal value added: total effect 2.3 2.2 2.5 2.4 2.4 gross output effect 2.5 2.5 2.6 2.8 2.6 intermediate input effect -0.2 -0.3 -0.1 -0.4 -0.2Banking and insuranceReal value added: total effect 1.3 1.3 1.3 1.5 1.3 gross output effect 1.6 1.4 1.4 2.2 1.9 intermediate input effect -0.3 -0.1 -0.1 -0.7 -0.6

170

For a more intuitive understanding, consider the example of computers and semiconductors.Assume that, initially, no price adjustment is present. Next, output prices of the computer industryare adjusted downward by 10 per cent and those of semiconductors by 5 per cent. What happens toreal intermediate inputs in the computer industry? The absolute measure of real semiconductor inputswill rise, given the price adjustment of semiconductors. However, at the same time, real output of thecomputer industry is also adjusted upwards, and by more than the measure of real semiconductorinputs (in addition, intermediate inputs are weighted with their respective input coefficient which, bydefinition is always less than unity). As a result, the measure of real intermediate inputs per unit ofreal gross output falls (a relatively smaller quantity of combined intermediate inputs is measured perunit of quality-adjusted gross output), and the measure of real value added per unit of real gross

output rises. But ~θ j is just the inverse of this relationship, namely real gross output over real value

added. Hence, ~θ j declines and reduces the weight on real gross output while nominal weights

remain unaffected.

Next, it is interesting to examine the effects on other sectors not included in the list of four ICT-related industries. By construction, price adjustment in ICT-related industries has a negative effect onthe measure of real value added in other sectors. Measures of real gross output do not change butmeasures of real intermediate inputs from ICT-related industries rise, in absolute terms as well asrelative to real gross output. The new measure shows that more intermediate inputs are needed in realterms to produce one unit of output; turned around, this implies a fall in real value added per unit ofgross output. Table 4 shows the changes in measured real value added of downstream industries ofthe four ICT sectors. The impact can be significant and is most accentuated in a number ofmanufacturing industries, in particular shipbuilding, aircraft, electrical and non-electrical machineryand professional goods. On average, and for the industries shown inTable 4, measured growth ratesof real value added (and therefore labour productivity) fell by 0.7 percentage points per year.

What determines the size of the impact on measures of downstream industries’ value added?First, the more important the purchases of intermediate inputs from ICT industries, the more thequality adjustment of prices of these inputs will be felt in the total measure of real intermediateinputs. A look at the input coefficients of the shipbuilding industry, for example, shows that they arein fact high compared to many other industries. This may be due at least partly to the fact that manyICT products (for example computers or navigation equipment installed on a ship, communicationequipment, telephone networks, etc.) are accounted for as an intermediate input to the shipbuildingindustry while they would most certainly qualify as investment goods in other industries.

Another point is the role of unadjusted prices: industry j’s real intermediate inputs per unit ofreal output depend on nominal input coefficients and on the relative changes of output and inputprices. If prices of intermediate inputs fall (because they are quality adjusted) while (unadjusted)output prices of the shipbuilding industry rise rapidly, this will further worsen the measured ratiobetween real intermediate inputs and real gross output. The unadjusted price index of theshipbuilding industry has, in fact, risen by more than those of other industries and so contributed tothe strong effects on measured real value added in the shipbuilding and ship-repairing sector.

171

Table 4. Quality adjustment of prices in ICT-related industries: 10 most affected sectorsDifference between annual average growth rates of adjusted and unadjusted real value added

Percentage pointsCanada A* NA* Germany A* NA*Shipbuilding & repairing -3.4 -2.8 Shipbuilding & repairing -1.7 -1.3Professional goods -1.1 -1.0 Aircraft -0.7 -0.6Petroleum & coal products -0.9 -0.3 Non-electrical machinery -0.6 -0.5Other manufacturing -0.4 -0.3 Professional goods -0.6 -0.5Motor vehicles -0.2 -0.4 Motor vehicles -0.5 -0.4Other producers -0.2 -0.2 Restaurants & hotels -0.3 -0.2Electrical apparatus, nec -0.2 -0.2 Mining & quarrying -0.3 -0.3Community, social & personal services -0.2 -0.2 Non-ferrous metals -0.3 -0.1Wholesale & retail trade -0.2 -0.2 Construction -0.3 -0.2Real estate & business services -0.2 -0.2 Wood products & furniture -0.2 -0.2

Japan A* NA* United States A* NA*Shipbuilding & repairing -0.8 -0.4 Professional goods -0.9 -0.7Professional goods -0.7 -0.5 Electrical apparatus, nec -0.8 -0.4Other manufacturing -0.6 -0.3 Aircraft -0.5 -0.6Electrical apparatus, nec -0.5 -0.5 Other manufacturing -0.3 -0.2Transport & storage -0.5 -0.2 Non-ferrous metals -0.3 -0.4Motor vehicles -0.4 -0.3 Motor vehicles -0.2 -0.2Other transport -0.4 -0.2 Community, social & personal services -0.1 -0.1Mining & quarrying -0.2 -0.1 Petroleum & coal products -0.1 -0.1Non-electrical machinery -0.2 -0.2 Paper, paper products & printing -0.1 -0.1Real estate & business services -0.2 -0.1 Real estate & business services -0.1 -0.1

United Kingdom A* NA*Electrical apparatus, nec -1.6 -0.6Professional goods -1.3 -1.0Aircraft -1.2 -0.8Shipbuilding & repairing -0.9 -0.5Motor vehicles -0.8 -0.3Electricity, gas & water -0.6 -0.2Non-electrical machinery -0.5 -0.3Other transport -0.4 -0.3Real estate & business services -0.4 -0.2Industrial chemicals -0.3 -0.2

A*: Additivity imposed.NA*: No additivity imposed.Source: OECD, STI/EAS.

172

Aggregation: real and nominal shares

Aggregation with additivity. As before, we distinguish between two methods to capture theeffects of quality adjustment on aggregate value added, depending on the underlying methodology foraggregation. The first method, which is standard practice in national accounts statistics, imposesadditivity and thus simply sums up the levels of sectoral real value added to obtain a measure ofaggregate real value added: this permits, for instance, the analysis of structures at constant prices andthe comparisons of levels of value added at constant prices in a framework that ensures consistency

between subsectors and parent industries. Thus, VA~

, the level of aggregate value added at constant

prices, is defined as VA VAjj

~ ~= ∑ , the sum of value added by industry and the associated growth rate

is simply ∆ ln~

VA . Similar to our earlier treatment, we distinguish between adjusted and unadjustedvariables and capture the effects of quality-adjusted price in ICT industries on aggregate measures ofreal value added by the expression:

∆ ∆ ∆ ∆ln~

ln~

ln~

ln~

VA VA VA VAa uja

jju

j

− =

−

∑ ∑ [10]

As before, the right-hand side of equation [10] can be approximated in discrete time by aTörnqvist-type index:

( ) ( )∆ ∆ ∆ ∆ln~

ln~ ~ ( ) ~ ( ) ln

~ ~ ( ) ~ ( ) ln~

VA VA s t s VA s t s VAa uja

ja

jja

ju

ju

jju− ≅ + − +∑ ∑τ τ1

21

2 [11]

Again, the point that emerges is that this is an aggregation method with weights ~ ~/

~s VA VAj j=

at base-year prices as they reflect each industry’s real share in aggregate value added.

Aggregation without additivity. The alternative method that does not impose additivity startsagain from a basic accounting principle: aggregate value added at current prices equals the sum of

sectoral value added at current prices. This is spelled out in [12], where p VAv denotes current-price

aggregate value added, the product of an aggregate price index pv and aggregate real value added

VA ; similarly, p VAjv

j stands for sectoral current-price value added with a price and a volume

component:

p VA p VAvjv

jj

= ∑ [12]

Differentiation with respect to time yields an expression that decomposes the growth rate ofnominal value added into a Divisia price index (first expression on the right-hand side of [13] and aDivisia quantity index of aggregate value added (second expression on the right-hand side of [13].The quantity index is similar to the quantity index in [10] with one important difference: sectoralrates of real value added are aggregated with nominal shares s j , not real shares ~s j .

& & & &p

p

VA

VAs

p

ps

VA

VAwith s

p VA

p VA

v

v jjv

jv j

j

jj

jv

j

vjj

+ = + =∑∑ [13]

The quantity indices in [13] can be rewritten in its discrete approximation as a Törnqvist-typeindex between two time periods t and τ . We then use this formulation to evaluate ∆ ln VA :

173

( )∆ ∆ln ( ) ( ) lnVA s t s VAj jj

j= +∑ τ 12 [14]

Finally, adjusted and unadjusted variables are introduced to assess the effects of sectoral price

adjustment on GDP. Let ∆ ln VAa and ∆ ln VAu be the quantity indices of real GDP, based onnominal sector shares and adjusted or unadjusted sectoral prices. Then, the difference

∆ ∆ln lnVA VAa u− represents the aggregate effect of sectoral price adjustment, using nominalshares for aggregation:

( )∆ ∆ ∆ ∆ln ln ( ) ( ) ( ln ln )VA VA s t s VA VAa uj j

jja

ju− = + −∑ τ 1

2 [15]

Aggregate value added: results

Effects on aggregate real value added of the quality adjustment of price indices in ICT-relatedindustries are displayed in Figures 3 and 4 according to the underlying aggregation procedure(additivity and non-additivity). The following points emerge from these graphs:

Figure 3. Effects of quality adjustment of prices in ICT-related industriesAggregation method: real shares (additivity imposed)

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

Germany Japan U.K. Canada USACha

nge

in a

nnua

l ave

rage

gro

wth

of

aggr

egat

e va

lue-

adde

d :

per

cent

age

poin

ts

Banking and insurance: -1% point

Communication services: -2% points

Communication equipment: -5% points

Office machinery and computers: -10%points

174

Figure 4. Effects of quality adjustment of prices in ICT-related industriesAggregation method: nominal shares (no additivity imposed)

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

Germany Japan U.K. Canada USACha

nge

in a

nnua

l ave

rage

gro

wth

of

aggr

egat

ere

al v

alue

-add

ed:

per

cent

age

poin

ts

Banking and insurance: -1% point

Communication services: -2% points

Communication equipment: -5% points

Office machinery and computers: -10%points

First, overall effects are visible, but not overwhelming. According to these results, the realgrowth rate of aggregate real value added was under-measured by anything between 0.2 percentagepoints per year in Canada to 0.6 percentage points in Japan. A comparison of these figures with theunderlying unadjusted growth rates confirms that the two are roughly in proportion: unadjusted realgrowth rates were relatively high in Japan and the United Kingdom (over 5 per cent) – the twocountries with the largest adjustments; while growth in Canada and the United States was only about3 per cent – in line with the lower adjustment results. Germany does not quite fit the picture: its(unadjusted) aggregate growth rate of about 4 per cent was clearly higher than that of Canada or theUnited States, while the order of magnitude of under-measurement is very much in line with thesetwo countries. However, this may be due to an inappropriate adjustment of Germany'scommunication equipment sector11.

Second, results in Figure 3 were obtained with the aggregation procedure that assumes additivitybetween levels of constant-price variables and is therefore based on real shares. Figure 4 shows thesame effects but based on the aggregation method that uses nominal shares. As it turns out, theaggregation procedure based on nominal sectoral shares produces a smaller deviation from unadjustedaggregate growth. This is to be expected as:

♦ Real shares for ICT-related industries increase over time even in the absence of specific qualityadjustment of prices. For example, the real (unadjusted) share of Japan’s computer industry inGDP rose from 0.77 per cent in 1985 to 0.9 per cent in 1990. Nominal shares of the sameindustry remain more or less constant; in the Japanese example they even decreased slightly to0.75 per cent.

♦ Real shares for ICT-related industries increase even more when prices are adjusted for qualitychanges. This causes a magnification of the real growth measure of these sectors that in turnraises the measure of aggregate real value added growth. To stay with the example of Japan,the 1990 real share of its computer industry rises from 0.9 per cent to 2.4 per cent when quality-adjusted prices replace conventional price indices, nearly tripling the contribution to measuredGDP growth. Nominal shares, on the other hand, are unaffected by quality adjustments ofprices and do not magnify the increased real growth measure of ICT-related industries. Thus, a

175

stronger impact should be expected under an aggregation method based on real shares thanunder one based on nominal shares.12

Conclusions

The present chapter examined the effects of an adjustment of sectoral prices for quality changeson sectoral and aggregate measures of real value added. Four principal conclusions arose from thisanalysis:

♦ First, the quality adjustment of gross output prices in ICT-related industries has a non-negligible impact on measured real value added, and therefore on measured labourproductivity, both in ICT-related industries and in other sectors of the economy. The size of theeffect on other sectors hinges on the extent to which products of ICT-related industries are usedas intermediate inputs in downstream sectors. While there are differences across the countriesexamined, there is also a common pattern: in each of the countries, manufacturing industries’measures of real value added tended to be more affected than service sectors’ measures.Shipbuilding, automobiles, professional goods and aircraft typically appear as those industriesfor which the downward correction of real value added measures is most pronounced.

♦ Second, the size of the effects on measures of industry real value added is greatly influenced bythe choice of the underlying methodology, specifically by the choice between nominal and realweights to combine the growth rates of variables. As the underlying question – should there beadditivity of levels of constant price variables? – remains open, our only conclusion is thatmethodology matters in particular at the industry level.

♦ Third, the impact on aggregate value added is visible, but not overwhelming. On an averageannual basis, growth rates of GDP of the five countries under consideration would have to beadjusted upwards by 0.2 to 0.6 percentage points over the period 1985-90 (and rates of inflationdownwards, correspondingly). Again, results may change if different weighting procedures areapplied to derive aggregate growth rates of real value added from sectoral growth rates.

♦ Fourth, it needs to be underlined that the quantitative effects derived in this study dependentirely on the assumptions about the degree of under-measurement of sectoral real grossoutput. While it could be argued that the assumed maximum deviations between adjusted andunadjusted price indices are too high, it should be kept in mind that we only evaluated theeffects of quality adjustment of prices in ICT-related industries, leaving aside the possibility ofunder-measurement in other sectors, for example the construction industry, transport servicesand the retail sector. Also, the period of observation is limited to the second half of the 1980s,not the time of the steepest decrease in prices of some ICT products, for example computers.The inclusion of earlier periods would, most likely, have revealed more pronounced effects atboth the sectoral and aggregate levels.

176

NOTES

1. At the same time it has been shown that there are sectors where hedonic price indices yield results that arehardly distinguishable from those derived under traditional deflators. A case in point are apparel priceindices (Liegey, 1994) where effects of hedonic adjustment have been weak and not uni-directional, despitethe potentially important role that hedonic-based quality adjustments play in an industry with largely“fashion-driven” goods that are subject to frequent product changes.

2. Wyckoff (1995) addressed the effects of quality adjustment of the office and computing machinery industryprices on productivity measures of that industry and its parent sectors: He showed that “differences in themeasurement of the change in output prices of the computer sector have a significant effect on the growthrate of labour productivity for this sector as well as the more aggregate industrial classification in which it isincluded (non-electrical machinery and fabricated metal products and equipment)”. These conclusions arederived from a sensitivity analysis whereby Wyckoff applied the US quality-adjusted price index of theoffice machinery and computers sector to current-price value added series of the same industry in otherOECD countries. He then compared the resulting real value added and labour productivity growth with thechanges in labour productivity based on conventional measures of value added.

3. In particular, it could be argued that computer-related service activities such as the creation of softwareprogrammes form part of ICT production. However, even if we accepted this notion, the empiricalclassification we have at our disposal for the current research is too broad to single out this service activitywhich is hidden in the very broad category of real estate and business services.

4. At this point our analysis meets the discussion about the Solow paradox, that is the apparent contradictionbetween a growing high quality ICT capital stock and the absence of an economy-wide pick-up ofproductivity growth. See Oliner and Sichel (1994) for a discussion of the size of the ICT capital stock.

5. In theory, the distinction between investment and intermediate goods cannot be grounded in the nature ofthe product, rather any product can be investment or intermediate good, – all that counts is whether the goodhas a service life that exceeds the accounting period, i.e. normally one year. A computer, bought andinstalled in the branch of a bank, for example, is clearly a capital good while the same computer, installed inthe navigation cabin of a ship, is an intermediate input to the shipbuilding industry.

6. For a discussion see Petit (1991).7. For a survey of methods to derive real value added measures in service industries of OECD countries, see

OECD (1996).8. For a presentation of the introduction of current-price chain weights in the calculation of the US GDP at the

end of 1995, see, for example, Landefeld and Parker (1995).9. In matrix form, the system of equations for real gross output by industry is solved as

[ ]q I p Ap p y= − − − −$ $ $1 1 1 where q is a vector of real gross industry output,$p is the diagonal matrix of

sectoral deflators, A is the matrix of current-price input coefficients and y is the vector of current price finaldemand.

10. There are alternatives to double-deflation that can be applied in an input-output context. Durand (1994)presents a method that consists of deflating industries’ direct and indirect contributions to final demanddeliveries (their value added by commodity) by the respective final demand commodity prices. Durandshows that this method features economic and statistical advantages in that it avoids, for example, negativereal value added and in that the quality of commodity price measurement tends to be superior to that ofproducer price measurement by industry. However, Durand’s approach requires matrices of intermediateuses by commodities and industries. The OECD Input-Output Database is confined to industry-by-industryintermediate flow matrices and precludes therefore the deflation via commodity prices as suggested byDurand.

11. See Section “Assumptions about mis-measurement”.12. A look at the five countries reveals that the impact of changing methodology are particularly pronounced in

the United Kingdom. Partly this may be due to the fact that the base year for the UK industry deflators is1980 (other countries 1985, 1986 and 1982), leading to a particular wide gap between the growth of

177

industries’ real and nominal shares in GDP. At the same time, the base year for the United States is alsoquite remote (1982) with little differential effects from the two methodologies while Japan, whose base yearis 1985, shows a sizeable difference in the results from the two aggregation procedures.

178

Annex

DATA SOURCES

All data for the present chapter are from The OECD Input-Output Database (OECD, 1995) which provides input-outputtables in comparable industry structure for 10 OECD countries (Australia, Canada, Denmark, France, Germany, Italy, Japan,Netherlands, the United Kingdom, the United States). The input-output tables are industry-by-industry tables that cover36 industries as shown in the table below. In the present analysis, producers of government services were excluded, as thetreatment of the government sector in input-output tables varies across countries.

Although 1985-90 is the reference period in the present study, this time span had to be adjusted as a function of theavailability of countries’ input-output tables: Canada: 1986-90, Germany: 1986-90, Japan: 1985-90, United Kingdom:1984-90, United States: 1985-90. Constant price matrices in the OECD Input-Output Database are expressed in prices ofdifferent base years: Canada: 1986, Germany: 1985, Japan: 1985, United Kingdom: 1980, United States: 1982. Althoughthe present analysis does not use constant price matrices as such, they provide the source for implicit deflators of gross outputby industry, the “unadjusted” prices indices used in the simulation.

OECD Input-Output Database: sector classificationNo ISIC Rev. 2 codes Description1 1 Agriculture, forestry and fishery2 2 Mining and quarrying3 31 Food, beverage and tobacco4 32 Textiles, apparel and leather5 33 Wood products and furniture6 34 Paper, paper products and printing7 351+352-3522 Industrial chemicals8 3522 Drugs and medicines9 353+354 Petroleum and coal products10 355+356 Rubber and plastic products11 36 Non-metallic mineral products12 371 Iron and steel13 372 Non-ferrous metals14 381 Metal products15 382-3825 Non-electrical machinery16 3825 Office and computing machinery17 383-3832 Electrical apparatus, nec.18 3832 Radio, TV and communication equipment19 3841 Shipbuilding and -repairing20 3842+3844+3849 Other transport equipment21 3843 Motor vehicles22 3845 Aircraft23 385 Professional goods24 39 Other manufacturing25 4 Electricity, gas and water26 5 Construction27 61+62 Wholesale and retail trade28 63 Restaurants and hotels29 71 Transport and storage30 72 Communication services31 81+82 Finance and insurance32 83 Real estate and business services33 9 Community, social and personal services34 Producers of government services35 Other producers36 Statistical discrepancy

179

REFERENCES

BAILY, M.N. and R.J. GORDON (1988), “The Productivity Slowdown, Measurement Issues, and theExplosion of Computer Power”, Brookings Papers on Economic Activity, 2.

COLE, R., Y.C. CHEN, J.A. BARQUIN-STOLLEMAN, E. DULBERGER, N. NELVACIAN andJ.H. HODGE (1986), “Quality Adjusted Price Indexes for Computer Processors and SelectedPeripheral Equipment”, Survey of Current Business, January.

DURAND, R. (1994), “An Alternative to Double Deflation for Measuring Real Industry Value-Added”, Review of Income and Wealth, Vol. 40, No. 3, September.

FLAMM, K. (1993), “Measurement of DRAM Prices: Technology and Market Structure”, inM. Foss, M. Manser and A. Young (eds.), Price Measurement and their Uses, University ofChicago Press.

GORDON, R.J. (1992), “Measuring the Aggregate Price Level: Implications for EconomicPerformance and Policy”, NBER Working Paper No. 3969, Cambridge, MA.

GRILICHES, Z. (1994), “Productivity, R&D, and the Data Constraint”, American Economic Review,March.

GRILICHES, Z. (1995), “Comments on Measurement Issues in Relating IT Expenditures toProductivity Growth”, Econon. Innov. New Techn., Vol. 3.

KURODA, M. and K. SHIMPO (1991), “Economic Growth in Japan 1960-1985”, paper prepared forthe Harvard-Japan Energy and Environment Workshop, Tokyo.

LIEGEY, P.R., Jr. (1994), “Apparel Price Indexes: Effects of Hedonic Adjustment, Monthly LaborReview, May.

LANDEFELD, J.S. and R.P. PARKER (1995), “Preview of the Comprehensive Revision of theNational Income and Product Accounts: BEA’s New Featured Measures of Output and Prices”,Survey of Current Business, July.

MILLER, R.E. and P.D. BLAIR (1985), Input-Output Analysis, Prentice Hall.MOTOHASHI, K. (1996), “ICT Diffusion and its Economic Impact on OECD Countries”, mimeo,

OECD, Paris.OLINER, S.D. and D.E. SICHEL (1994), “Computers and Output Growth Revisited: How Big Is the

Puzzle?”, Brooking Papers of Economic Activity, 2.OECD (1991), Technology and Productivity: The Challenge for Economic Policy, Paris.OECD (1995), The OECD Input-Output Database, Paris.OECD (1996), Services: Measuring Real Annual Value Added, Paris.PETIT, P. (1991), “New Technology and Measurement of Services: The Case of Financial

Activities” in OECD, Technology and Productivity: The Challenge for Economic Policy, Paris.POPKIN, J. (1992), “The Impact of Measurement and Analytical Issues in Assessing Industry

Productivity and its Relationship to Computer Investment”, study by Popkin and Companysubmitted to International Business Machines Corporation [quoted after Sichel (1994)].

SAKURAI, N., E. IOANNIDIS and G. PAPACONSTANTINOU (1996), “The Impact of R&D andTechnology Diffusion on Productivity Growth: Evidence for 10 OECD Countries in the 1970sand 1980s”, STI Working Paper 1996/2, OECD, Paris.

SICHEL, D.E. (1994), “The Computer Paradox and the Productivity Slowdown: Is Mis-measurementthe Culprit?”, unpublished paper, Washington, Brookings Institution.

STATISTICS CANADA (1996), “Measuring the Global Information Infrastructure”, unpublisheddocument presented to the OECD ICCP meeting (March)

180

TRIPLETT, J.E. (1990), “Hedonic Methods in Statistical Agency Environment: An IntellectualBiopsy”, in E.E. Berndt and J.E. Triplett (eds.), Fifty Years of Economic Measurement,Chicago.

WYCKOFF, A.W. (1995), “The Impact of Computer Prices on International Comparisons of LabourProductivity”, Economic Innovation New Technology, Vol. 3.